Let $X$ be a compact Hausdorff topological space, and $\mathcal C^0 (X) = \{f:X\to\mathbb{R}; \ f \text{ is continuous }\}$. It is well known that for any bounded linear functional $\phi: \mathcal C^0(X)\to\mathbb{R},$ such that $\phi(f)\geq 0$ if $f\geq 0$ ($\phi$ is called a positive linear functional), then there exists a unique regular Borel measure $\mu$, such that $$\phi(g) = \int g\ \mathrm d\mu, \ \forall \ g\in \mathcal C^0(X). $$ This result follows from a direct application of Riesz–Markov–Kakutani representation theorem.

If we drop the Hausdorff hypothesis (only assuming $X$ as compact topological space). Then we can lose the uniqueness of the measure that represents the linear functional. A famous example is the compact topological space "$[0,1]$ with to origins". In this case the functional $\phi: \mathcal C^0(X)\to\mathbb{R}$, $\phi(f) = f(0)$ can be written as $\int f\ \mathrm{d}\delta_0$ or $\int f\ \mathrm{d}\delta_{0'}.$

I would like to know if we still have the existence of a measure that represents the functional. In other words, I would like to know if the following theorem is true

Possible Theorem: Let $(X,\tau)$ be a compact non-Hausdorff space, and $\Lambda : \mathcal C^0(X)\to\mathbb{R}$ a positive bounded linear functional, then there exists a measure $\mu: \mathcal B(\tau)\to \mathbb{R}$ (where $\mathcal B(\tau)$ is the smallest $\sigma$-algebra such that $\tau\subset \mathcal B(\tau))$, such that $$\Lambda(f) = \int f\ \mathrm{d}\mu, \ \forall \ f\in \mathcal C^0(X).$$

Can anyone help me?

I have searched online but I was not able to find a result in the non-Hausdorff case.


1 Answer 1


The answer is yes.

First, it follows from the following result and the Riesz–Markov–Kakutani representation theorem that we can always find a suitable Baire measure representing a positive linear functional.

Theorem: Let $X$ be any topological space. Then there exists a completely regular Hausdorff space $Y$ and a continuous surjection $\tau:X\to Y$ such that the function $g\mapsto g\circ\tau$ is an isomorphism from $C_B(Y)$ onto $C_B(X)$.

This is Theorem 3.9 of "Rings of continuous functions" (1960) by Gillman and Jerison.

So the problem reduces to the question whether a Baire measure on a compact topological space can be extended to a Borel measure. We can do this using the following result, which specializes the very abstract Theorem 2.6.1 of "Convex Cones" (1981) by Fuchssteiner and Lusky.

Theorem: Let $X$ be a non-empty compact topological space and $L:\mathcal{C}^0_+(X)\to\mathbb{R}$ be an additive function on the cone of nonnegative continuous functions on $X$ such that $L(g)\leq\max g$ for all $g$. Then there exists a Borel probability measure $\nu$ on $X$ such that $$L(g)\leq\int g~\mathrm d\nu$$ for all $g\in \mathcal{C}^0_+(X)$.

For nonzero $\Lambda$, let $L=1/\Lambda(1)\cdot \Lambda$. Then the measure $\mu=\Lambda(1)\cdot\nu$ does the trick.

It should be noted that the resulting Borel measure need not be regular. For non-Hausdorff $X$, there is no point in going beyond Baire measures.

  • $\begingroup$ Just one question that I am not following (btw, the trick of inducing a Baire-measure using $\tau$ was really clever). Using the second theorem we are able to find a Borel probability measure such that $L(f) \leq \int f \ \mathrm{d}\nu$. Why this solve the problem? Since we are interested in the equality of both terms. $\endgroup$ May 17, 2020 at 18:09
  • 2
    $\begingroup$ @MatheusManzatto I wrote the statement with an inequality, so that it matches up easily with the statement in the book. Here is how one can get equality: Assume without loss of generality that $\Lambda(1)=1$ and $0\leq g\leq 1$ (the general case follows from rescaling). Then $\Lambda(g)\leq\int g~\mathrm d\nu$ and $\Lambda(1-g)\leq \int 1-g~\mathrm d\nu$ together with $1=\Lambda(1)=\Lambda(g)+\Lambda(1-g)\leq \int g~\mathrm d\nu+\int 1-g~\mathrm d\nu=1$ implies that $\Lambda(g)=\int g~\mathrm d\nu$. $\endgroup$ May 17, 2020 at 18:19
  • 1
    $\begingroup$ Nice. It seems to me from you answer and your comments that the Corollary to Theorem II.2.6.1 from the book of Fuchssteiner and Lusky does the job alone - it's even stated in the book as a Riesz representation theorem for non-Hausdorff compact topological spaces. Is the "Tychonoffication" result from Gilman-Jerison even needed? $\endgroup$ May 18, 2020 at 6:09
  • 1
    $\begingroup$ @PedroLauridsenRibeiro You are right. The first result is for context. $\endgroup$ May 18, 2020 at 6:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.